龔鵬程對話海外學(xué)者第六十七期：在后現(xiàn)代情境中，被技術(shù)統(tǒng)治的人類社會，只有強化交談、重建溝通倫理，才能獲得文化新生的力量。這不是誰的理論，而是每個人都應(yīng)實踐的活動。龔鵬程先生遊走世界，并曾主持過“世界漢學(xué)研究中心”。我們會陸續(xù)推出“龔鵬程對話海外學(xué)者”系列文章，請他對話一些學(xué)界有意義的靈魂。范圍不局限于漢學(xué)，會涉及多種學(xué)科。以期深山長谷之水，四面而出。

菲利普·M·尼科爾斯教授（Professor Philip M. Nichols）

美國賓夕法尼亞大學(xué)商業(yè)社會責(zé)任教授、法律研究與商業(yè)倫理學(xué)教授，美國國際法學(xué)會反腐敗法利益小組聯(lián)合主席

龔鵬程教授：您好。在您的論文《使用同構(gòu)的社會學(xué)理論來評估通過貿(mào)易制裁改變政權(quán)的可能性》中，您通過社會學(xué)理論來預(yù)測貿(mào)易制裁的結(jié)果。社會學(xué)對理解貿(mào)易制裁有什么幫助，您可以舉一些例子嗎？

菲利普·M·尼科爾斯教授：龔教授，您好。國家或某些政體是為了推進一些政策或目標(biāo)而去實施貿(mào)易制裁的。例如，世界大部分地區(qū)在南非推行種族隔離制度時對南非實施貿(mào)易制裁，這既是為了表達(dá)對種族隔離的譴責(zé)，也是為了推動南非走向變革。

事實上，制裁的實施通常是為了激勵特定國家/地區(qū)的規(guī)則變化。社會學(xué)的一個分支，稱為制度社會學(xué)，研究規(guī)則和制度為什么以及如何在政體中發(fā)生變化。

他們的研究闡明了一些例如路徑依賴性和有限知識（bounded knowledge）等有用的概念。

路徑依賴性描述了制度的強大慣性，除非政體經(jīng)歷了強大的變革推動，否則它們將會一直保持類似的模式。以黎巴嫩為例，即使他們知道在特定群體之間分配政府權(quán)力的系統(tǒng)不再有效，但還是繼續(xù)構(gòu)想類似形式的政府制度。

另一方面，有限知識（bounded knowledge）指出，即使對于外部觀察者來說有些制度的改變是有意義的，但是政體依舊不會采用他們不知道的制度。

有時，政體實施貿(mào)易制裁，是為了推動另一個政體使其制度遵照標(biāo)準(zhǔn)。例如，美國對古巴實施貿(mào)易制裁，以推動古巴“加入在西半球蓬勃發(fā)展的民主國家共同體”，換言之，使其制度與美洲國家組織成員國的制度一致。

制度社會學(xué)家將這種改變自身制度以看起來像其他制度的過程稱為‘同構(gòu)’。消極或積極的誘因都可以導(dǎo)致‘強制同構(gòu)’，使用積極誘因的最著名最成功的例子是向中歐和東歐新獨立的國家提供歐盟成員資格。加入歐盟需要將許多制度與歐盟標(biāo)準(zhǔn)保持一致，而這些國家也在創(chuàng)紀(jì)錄的時間內(nèi)完成了這些工作。

而貿(mào)易制裁，則是一種負(fù)面誘因。因此，當(dāng)貿(mào)易制裁的目的是推動一個政體使其制度符合某種標(biāo)準(zhǔn)時，社會學(xué)制度學(xué)家應(yīng)該對這些制裁的結(jié)構(gòu)有深刻的見解。

例如，為什么對南非的制裁最終成功，而對伊朗的制裁迄今都失敗了？什么時候?qū)€人（如在俄羅斯入侵烏克蘭后）的制裁，比那種對整個政體（如在委內(nèi)瑞拉可疑選舉與隨之而來的鎮(zhèn)壓后）的制裁更有效？ 這些問題屬于制度社會學(xué)家的范疇，可以為制裁的運作提供真正的見解。

Nations or other polities impose trade sanctions for the purpose of furthering some policy or objective. For example, much of the world-imposed trade sanctions on South Africa when it practiced apartheid, both to express condemnation of apartheid and to push South Africa towards change. Indeed, sanctions are often imposed in order to incentivize changes in the rules in a particular country. One branch of sociology, called institutional sociology, studies why and how rules – institutions – change within a polity. Their study has illuminated useful concepts such as path dependency and bounded knowledge. Path dependency describes the powerful inertia of institutions, which maintain similar form unless the polity experiences a powerful agent of change. Lebanese, for example, know that their system of allocating government powers among specified groups no longer works, but continue to envision government institutions in that form. Bounded knowledge, on the other hand, points out that polities cannot adopt institutions of which they have no knowledge, even if to an outside observer it would make sense to change in that way.

Sometimes, polities impose trade sanctions in order to push another polity to conform its institutions to a standard. The United States, for example, imposes trade sanctions on Cuba in order to push Cuba to “join the community of democratic countries that are flourishing in the Western Hemisphere,” in other words to align its institutions with those of the members of the Organization of American States. Institutional sociologists call the process of institutions changing to look like those of othersisomorphism; coercive isomorphism occurs because of negative or positive incentives. The most famous, and successful, example of the use of positive incentives is the offer of membership in the European Union to newly independent countries in Central and Eastern Europe. Accession to the European Union requires aligning myriad institutions with the EU’s standards, which those countries accomplished in record time.

Trade sanctions are a form of negative incentives. So, when the objective of trade sanctions is to push a polity to conform its institutions to some standard, sociological institutionalists should have substantial insights in how those sanctions should be structured. Why, for example, were the sanctions on South Africa ultimately successful, while those on Iran have so far failed? When are sanctions on individuals, of the sort that followed Russia’s invasion of Ukraine, more effective than sanctions on the entire polity, of the sort imposed on Venezuela following a questionable election and subsequent repression? These are the sorts of questions that fall within the field of institutional sociologists, and that could provide true insights into the working of sanctions.

龔鵬程教授：嚴(yán)重的腐敗，損害著公共和私人組織的利益。這個問題在新興經(jīng)濟體中似乎很嚴(yán)重，因為它們正在經(jīng)歷快速的變化，所以這使它們更容易受到腐敗的影響。在您的論文《通過打擊腐敗創(chuàng)造共享價值》中，您提出了“創(chuàng)造共享價值”的戰(zhàn)略，來提高盈利能力。這個商業(yè)戰(zhàn)略需要什么？公司如何實施它？您能舉一些例子嗎？

菲利普·M·尼科爾斯教授：創(chuàng)造共享價值戰(zhàn)略 (CSV) 是由邁克爾·波特和他的同事命名并廣為人知，但這個概念在波特之前便已出現(xiàn)。CSV 并非慈善，而是一種提高商業(yè)公司底線的策略。

簡而言之，該戰(zhàn)略認(rèn)識到商業(yè)公司不是在真空中運營，而是深深植根于其周圍的社會結(jié)構(gòu)和制度中：企業(yè)就像花盆里的植物、池塘里的魚。如果植物想要保持健康和生長，就必須改善土壤質(zhì)量并更換花盆。如果魚開始生長，它將需要更好的水質(zhì)和更大的池塘。同樣，如果企業(yè)想要保持健康和成長，它可能需要改善其運營所在的社會環(huán)境。

其實有很多公司使用CSV 策略的例子，只是很少在北美出現(xiàn)。

我最想提到的是食品公司Aace Foods的例子。一對曾在國外工作的尼日利亞夫婦回國后成立了投資基金并收購了Aace，隨后發(fā)現(xiàn)它所依賴的當(dāng)?shù)剞r(nóng)業(yè)產(chǎn)業(yè)基礎(chǔ)設(shè)施不合格、技術(shù)不可靠且知識過時。他們沒有像一些顧問建議的那樣將業(yè)務(wù)轉(zhuǎn)移到東歐，而是著手改善尼日利亞的農(nóng)業(yè)。

他們幫助建立了一些金融項目以平衡農(nóng)業(yè)周期并為改進技術(shù)獲得信貸。他們幫助建立了合作社以增強農(nóng)民的商業(yè)力量，同時還傳播最新的知識與實踐。所有農(nóng)民都可以獲得這些幫助（但將農(nóng)民與Aace 綁定是不可能的），因此他們的生活全都得到了提升。Aace 的運營和收入也是如此，現(xiàn)在成為了一家蓬勃發(fā)展的食品公司。

腐敗，對社會和全球都會造成重大損害，也給個體商業(yè)公司帶來了實際成本。腐敗降低商業(yè)成本的神話已被徹底推翻。

因此，很明顯，一家商業(yè)公司可以通過改善其經(jīng)營所在的社會環(huán)境來提高其地位，這種模式也就是努力減少腐敗?，F(xiàn)在越來越多的商業(yè)公司開始相互合作，為世界各地的反腐敗做出貢獻(xiàn)。CSV有助于解釋原因，并可能根據(jù)其他領(lǐng)域的CSV 經(jīng)驗提供有用的指導(dǎo)。

The Creating Shared Value strategy (CSV) was named and made famous by Michael Porter and his colleagues, but long predates Porter. CSV is not charitable, it is instead a strategy for improving the bottom line of a business firm. Put very simply, the strategy recognizes that a business firm does not operate in a vacuum, but instead is deeply embedded in social structures and institutions around it. A business firm is like a plant in a pot, or a fish in a pond. If a plant wants to stay healthy and grow, the soil must be improved and the pot changed. If the fish starts to grow, it will need better water and a bigger pond. Similarly, if a business wants to stay healthy and grow, it might need to improve the social context in which it operates.

There are lots of examples of the use of the CSV strategy, very few of which took place in North America. One of my favorites is Aace Foods. A Nigerian couple who had worked abroad set up an investment fund when they returned to Nigeria; the fund bought Aace and found itself dependent on an agricultural industry with substandard infrastructure, unreliable technology, and dated knowledge. Rather than move their operations to Eastern Europe, as some consultants suggested, they undertook to improve agriculture in Nigeria. They helped establish financial facilities to even out agricultural cycles and to secure credit for improved technology. They helped create cooperatives, to boost the commercial power of farmers. They helped disseminate up-to-date knowledge and practices. These improvements were available to all farmers (tying farmers to Aace would have been impossible), and so the lives of all farmers were lifted. So were the operations and revenues of Aace, which is now a thriving foodstuffs company.

Corruption inflicts significant damage on society and on our planet. Corruption also imposes real costs on individual business firms. The myth that corruption reduces business costs has been soundly disproven. Thus, it seems obvious that a business firm could improve its position by improving the social context in which it operates, in the form of working to reduce corruption. Business firms are cooperating with one another and contributing to anticorruption around the world in increasing numbers; the CSV strategy helps to explain why and might provide useful guidance based on CSV experiences in other arenas.

龔鵬程教授：控制腐敗的一種方法，是在組織中營造一種濃厚的道德環(huán)境。在您的論文《最大化利益相關(guān)者信任作為控制腐敗的工具》中，您建議使用可衡量的利益相關(guān)者信任作為公司道德文化的代表。您能解釋一下這種方法是如何工作的嗎？

菲利普·M·尼科爾斯教授：“道德環(huán)境”一詞描述了組織文化中處理對與錯以及應(yīng)該與不應(yīng)該的問題。因為為文化制定一個衡量標(biāo)準(zhǔn)其實是一個非常困難的挑戰(zhàn)，就像衡量愛情一樣，所以我們可以轉(zhuǎn)向?qū)ふ铱珊饬康拇怼?/p>

然而，使用錯誤的代表可能會導(dǎo)致結(jié)果不準(zhǔn)確甚至產(chǎn)生風(fēng)險。就準(zhǔn)確性而言，文化是異質(zhì)的并因地而異。例如，如果我們以一家商業(yè)公司的成員拒絕行賄要求的次數(shù)作為代表，那么我們該如何比較一家面對100次行賄要求中拒絕了99次的公司和一家從未面臨行賄要求的公司？就風(fēng)險而言，使用錯誤的白標(biāo)會引發(fā)對古德哈特定律的擔(dān)憂：使用代表會導(dǎo)致人們將該代替視為目標(biāo)甚至操縱該代表，并且無視其他行為。

作為個體，我們應(yīng)以多種方式展現(xiàn)我們的社會成熟和善良，商業(yè)公司也應(yīng)如此。

過分強調(diào)一兩個行為而忽視其他行為，可能導(dǎo)致用有害的方式扭曲該行為。例如，如果一家商業(yè)公司的高道德文化的代表是拒絕賄賂要求，那么它既可能會將自己置于需要賄賂的境地，同時也可能會忽略其他幾十個同樣應(yīng)該成熟、負(fù)責(zé)任地行為處事的方面。

使用利益相關(guān)者信任作為代表，則可能可以規(guī)避這類風(fēng)險。利益相關(guān)者是與組織互動或受組織影響的個人和實體，信任是一種行為者在有風(fēng)險的情況下會表現(xiàn)合宜的感覺。信任是可以通過反復(fù)成功的互動來建立的（而聲譽在很多情況下都是類似情形的其他人的報告）。

有趣的是，人們認(rèn)為合宜的行為，與我們認(rèn)為的一般道德行為有很大的重疊。因此，信任感不依賴于特定的行為，而是依賴于適合該關(guān)系和該背景的行為，但仍屬于反映強烈道德氛圍的行為類別。這種概括，意味著我們避免了伴隨特定行為選擇的不準(zhǔn)確性，以及過分強調(diào)或玩弄該行為的危險。利益相關(guān)者的信任，似乎是衡量道德氛圍強度的有用指標(biāo)。

The term “ethical climate” describes that part of an organizational culture that deals with questions of right and wrong and should and shouldn’t. Developing a metric for culture presents extraordinary challenges (it is sort of like measuring love), so we turn to the use of proxies. The use of the wrong proxy, however, can be inaccurate or dangerous. In terms of accuracy, cultures are idiosyncratic and adapted to local contexts. If we use as a proxy, for example, the number of times that members of a business firm refuse demands to pay a bribe, then how do we compare a firm in which people refused ninety-nine out of one hundred demands to pay a bribe to a firm that has never faced a demand to pay a bribe? In terms of danger, the use of the wrong proxy raises concerns about Goodhart’s law: that the use of a proxy causes people to target or even game that proxy, and disregard other behaviors or acts. As individual people we manifest our social maturity and our goodness in myriad ways; the same is true of business firms.

Overemphasizing one or two acts and ignoring the others could distort behavior in harmful ways. For example, if the proxy for a strong ethical culture were refusing bribe demands, business firms might put themselves in situations in which bribes would be demanded, but might ignore the dozens of other ways in which they should behave maturely and responsibly.

The use of stakeholder trust as a proxy might avoid these dangers. Stakeholders are people and entities that interact with or are affected by an organization; trust is a sense that an actor will behave appropriately under conditions of risk. Trust seems to be created through repeated successful interactions (and reputation is in many ways the reports of others regarding the same). Interestingly, what people think of as appropriate behaviors strongly overlaps with what we think of as ethical behavior in general. Thus, a sense of trust is not dependent on a particular behavior but instead on the behaviors that are right for that relationship and that context, but that still fall within the category of behaviors that reflect a strong ethical climate. This generalization means we avoid the inaccuracies that accompany the choice of a specific behavior, as well as the danger of overemphasizing or gaming that behavior. Stakeholder trust seems to be a useful proxy for measuring the strength of an ethical climate.

龔鵬程教授：您最近寫了幾篇關(guān)于機器學(xué)習(xí)算法（machine learning algorithms）腐敗的文章，例如，您的論文《賄賂機器：在革命開始時保護算法的完整性》就深入談及了這個問題， 這很有趣。，看來機器學(xué)習(xí)也存在腐敗的可能性。但是，算法究竟是如何被破壞的呢？ 社會可以做些什么來控制這種腐敗？

菲利普·M·尼科爾斯教授：算法是人工構(gòu)造的，而不是自然產(chǎn)物。 算法由處理數(shù)據(jù)的規(guī)則組成（如果 x，則 y——如果從外部監(jiān)視器流入的數(shù)據(jù)表明亮度已降至 80 流明以下，則完成電路以打開燈）。通過收集足夠數(shù)量的數(shù)據(jù)并使用簡化模型解釋該數(shù)據(jù)來創(chuàng)建算法。

交通傳感器可能會收集幾個月的數(shù)據(jù)，然后使用簡化模型解釋數(shù)據(jù)，該模型表明數(shù)據(jù)中的每個積極跡象都代表一輛車，所有車輛的目標(biāo)是盡可能快地移動從一點到另一點。 不是所有的積極跡象都是車輛，一大群人可能會在傳感器上注冊，并不是所有的車輛都想快速移動，有些可能是觀光，有些只是開車，但我們需要簡化模型 數(shù)據(jù)感。

這些示例使用非常簡單的算法。實際上，現(xiàn)在許多重要的決策都是通過算法做出的，而且很快大多數(shù)決策都會使用非常復(fù)雜的規(guī)則集來處理數(shù)據(jù)。 與今天的非算法決策一樣，數(shù)十億和數(shù)萬億歐元將在這些算法決策的基礎(chǔ)上流動。

關(guān)于交通流量的決定會影響交付公司的成本和收入，關(guān)于許可的決定會影響建設(shè)范圍，關(guān)于醫(yī)療保健的決定會影響藥品和醫(yī)療設(shè)備的盈利能力。會有很多誘惑，特別是在已經(jīng)經(jīng)歷地方性腐敗的建筑行業(yè)等領(lǐng)域，破壞將做出這些決定的算法。

有幾種方法可以故意破壞算法。用于創(chuàng)建算法的數(shù)據(jù)可能存在偏差（例如，快遞公司可能會關(guān)閉一些傳感器并反復(fù)駛過其他傳感器）。數(shù)據(jù)的存儲和檢索，可能會出現(xiàn)偏差（對競爭對手有利的數(shù)據(jù)可能會被降低使用的可能性，就像商業(yè)公司試圖在出現(xiàn)在搜索引擎中獲得優(yōu)于競爭對手的優(yōu)勢一樣）。使不同數(shù)據(jù)具有可比性的方式可能會出現(xiàn)偏差（有關(guān)睡眠的數(shù)據(jù)可能優(yōu)先于有關(guān)飲食的數(shù)據(jù)）。

這些都是可能的，并且可能已經(jīng)在某種程度上發(fā)生了。更有可能的是，設(shè)計算法的人可能會被賄賂使用傾斜的簡化模型（我們已經(jīng)在簡化模型時遇到了有害偏見的問題）或編寫有利于一個參與者而不是另一個參與者的規(guī)則。

這似乎有些牽強，但在制藥等行業(yè)，商業(yè)公司目前經(jīng)常賄賂醫(yī)生開藥，即使這些藥物對患者有害或患者不需要它們。似乎不可避免的是，他們還將試圖賄賂設(shè)計算法的人，這些算法將有助于大規(guī)模處方?jīng)Q策。

在當(dāng)前規(guī)則下，很難檢測到算法的損壞。大多數(shù)算法都以商業(yè)機密的形式免受公眾審查。 這些規(guī)則需要改變。伴隨著數(shù)字革命的社會發(fā)展中也出現(xiàn)了許多深刻變革，法律法規(guī)也將在未來幾十年發(fā)生很大變化。 在這些變化中，我們必須確保法律能夠讓我們對算法進行審查和評估，這些算法將做出影響我們生活質(zhì)量和地球狀況的決策。

Algorithms are human constructs, not natural artifacts. Algorithms consist of rules for processing data (ifx, then y – if the data flowing in from the external monitor indicates that the brightness has fallen below 80 lumens, then complete the circuit to turn on the lights). Algorithms are created by gathering a sufficient amount of data and interpreting that data using simplifying models. Traffic sensors might gather data for a few months, and then that data would be interpreted using a simplifying model that suggests that every positive indication in the data represents a vehicle, and that the goal of all of the vehicles is to move as quickly as possible from one point to another. Not all of the positive indications are vehicles, a large group of people might register on the sensor, and not all vehicles want to get around quickly, some might be sightseeing and some just out for a drive, but we need the simplifying model to make sense of data.

These examples use very simple algorithms. In reality, many important decisions are now made algorithmically and soon most will be, using very complex sets of rules for processing data. As with non-algorithmic decisions today, billions and trillions of euros will move around on the bases of these algorithmic decisions. Decisions about traffic flow affect the costs and revenues of delivery companies, decisions about permits affect the scope of construction, decisions about healthcare affect the profitability of pharmaceuticals and medical devices. There will be a lot of temptation, particularly in fields such as the construction industry which already experience endemic corruption, to corrupt the algorithms that will make those decisions.

There are several ways in which algorithms could be intentionally corrupted. The data used in creating the algorithm could be skewed (a delivery company could, for example, black out some sensors and repeatedly drive over other sensors). The storage and retrieval of data could be skewed (data beneficial to a competitor could be made less likely to be used, in the same way that business firms attempt to gain advantage over competitors in appearing in search engines). The ways that disparate data are made comparable could be skewed (data about sleep could be given priority over data about diet). These are all possible and probably already happen to some degree. Even more likely, the people who design algorithms could be bribed to use skewed simplifying models (we are already experiencing problems with harmful bias in simplifying models) or to write rules that favor one actor over another. This may seem far-fetched, but inoindustries such as pharmaceuticals, business firms currently frequently bribe doctors to prescribe their drugs, even when those drugs harm patients or patients do not need them. It seems inevitable that they will also try to bribe people who design algorithms that will assist in prescription decisions on a mass scale.

Under current rules, it would be very difficult to detect the corruption of algorithms. Most algorithms are shielded from public scrutiny, in the form of trade secrets. Those rules need to change. Laws and regulations are going to change a lot over the next few decades, as society goes through the profound changes that accompany the digital revolution. Among those changes, we must ensure that the law enables our scrutiny and evaluation of the algorithms that will make decisions that shape the quality of our lives and the condition of our planet.

龔鵬程，1956年生于臺北，臺灣師范大學(xué)博士，當(dāng)代著名學(xué)者和思想家。著作已出版一百五十多本。

辦有大學(xué)、出版社、雜志社、書院等，并規(guī)劃城市建設(shè)、主題園區(qū)等多處。講學(xué)于世界各地。并在北京、上海、杭州、臺北、巴黎、日本、澳門等地舉辦過書法展?，F(xiàn)為中國孔子博物館名譽館長、美國龔鵬程基金會主席。